AI: Distinguishing Between Hard Facts And Science Fiction
Artificial intelligence (AI) is beginning to exert a powerful influence over humanity in a myriad of ways. The possibilities seem endless. But, as with all great opportunities, AI brings immense risks, many of which we have seen glamorised in science fiction.
Few understand what AI actually is and how it will shape our society, our work and our daily lives over the short and medium term
At Origin, we see big opportunities for AI in financial services. But we’re also conscious that there’s so much hyperbole around the subject it isn’t always easy to know what these opportunities are. It’s important we fully grasp what AI is before we can speculate as to the implications for our industry.
* * * * *
AI today is more strictly called narrow AI or weak AI.
Simply, this means it performs a narrow, defined task. Air traffic control systems, driverless cars, smartphones and, increasingly, financial markets, are all run by AI algorithms. These systems are designed by man and computers are then trained and put to task. They work for sustained, error-free periods within a system’s parameters, subject to human monitoring and oversight.
Scientists and researchers are working to develop Artificial General Intelligence (AGI), which allows for the possibility that machines will begin to outperform humans, operating without supervision, designing systems of their own. Singularity is the point at which AI will abruptly trigger runaway technological growth. This is when computers start building computers, and it’s this point that we should be most concerned about.
Many experts have come out in support of AGI, and many have spoken about its dangers. Elon Musk is on record stating, “AI is a fundamental risk to the existence of civilisation.” This is true in theory, but some way off. The more near real-world implications of AI – specifically, narrow AI – should be our primary concern and the longer term, science fiction-inspired outcomes should be saved for another day (or decade).
Putting science fiction aside, we see that so much of what is being done by AI right now is good, progressive but, in short, no harder to comprehend than machine learning, whereby machines repetitively use systems and parameters that humans have built over and over again. This is a form of regression analysis, just another strand of statistics. AI computers are able to do regressions faster than humans ever could, but the computer still needs the human to point it in the right direction.
This form of AI has led to some extraordinary outcomes. Humans haven’t beaten a computer at chess since 1997, when Deep Blue beat Gary Kasparov for the first time. In 2011, Watson beat human players on the language based game show Jeopardy! In 2015, heads-up Hold’em poker was “fully solved” by Cepheus. This year, Deepmind’s AlphaGo retired from competitive Go after defeating the human world number one 3-0.
These are all still examples of AI in a narrow sense, as self learning models, whereby the moves, rules and outcomes of a game can be learnt quickly by a supercomputer and the play optimised. This still falls short of AGI, which remains a long way off.
However, that’s not to say that narrow AI, using sophisticated machine learning and self-learning models, isn’t having a huge impact on the world, particularly in our world of finance. Our industry offers so much data for models to train against. Hence, the advancements that AI offers should be great.
Finance is no stranger to this sort of progress. Quantitative analysis has been the bedrock upon which much of finance – especially hedge funds – has been built over the past three or four decades. Narrow AI should be considered the next step along this path.
A good corollary of how this progress will play out is biology. In biology, many recent breakthroughs have come from machine learning technology, which is used to “train” computer models via the massive amount of genetic data that can be created now that it’s becoming cheaper to sequence DNA. As a result, all genetic testing being done – such as tests to see if you’re predisposed to diseases, or the tracking of your gene heritage – is based on machine learning.
You can imagine finance benefiting from this approach, with machine learning being employed to analyse huge amounts of market data in a quicker, more efficient and optimised way. This will lead to rich insights for investors to act upon. Financial firms are already using AI to prepare and analyse vast amounts of social media entries, accounting data, sales histories and forecasts.
And this feels like just the beginning. The holy grail for AI in finance is for it to find investment opportunities of its own hidden in data without human assistance. Like Deep Blue finding the checkmate move, this approach is narrow, but its implications for finance will be huge.
Ultimately, the true test of AI and whether it is adopted more widely by the financial community will be the same as any other innovation from thousands of years of financial development – it will be judged by its ability to make money.
If its adoption by large financial institutions is profitable, then we see no reason why AI won’t become a core part of the future of finance.